For diffusion-weighted MR imaging of water in neural tissue, it has been well established that the attenuation signal is non-monoexponential indicating non-Gaussian dynamics within the bulk microstructure. However, since water is non-specific to any particular biological compartment, it is challenging to disentangle the structural origins that give rise to the complex diffusion processes. Alternatively, diffusion-weighted MR spectroscopy (DWS) allows for relatively compartment-specific analysis of microstructure. Previously, at ultra-short diffusion times (<10 ms), non-Gaussian diffusion was established in the whole rat brain, providing sensitivity to the regime when the metabolites transition from free to hindered diffusion¹. However, a monoexponential signal decay was assumed to estimate the diffusion coefficient (D), due to low spatial resolution of the diffusion-weighting (b)¹. Here, we use a long diffusion time and high spatial resolution to establish non-Gaussian metabolite displacement in white matter (WM) and gray matter (GM) volumes of interest (VOIs).

Recently, it has been shown that water molecules exhibit Non-Gaussian subdiffusive dynamics according to the following attenuation model,

$$S(\mathbf{b})/S(0)=E_\alpha(-\mathbf{b}D), \,\,\,\,\,\,\,\, \,\,\,\,\,\,\,\,[1]$$

where E_α is the Mittag-Leffler function, which is the characteristic function derived from anomalous transport theory^2,3. When 0<α<1, the dynamics are subdiffusive and when α=1, the diffusion dynamics are Gaussian and the MLF becomes the exponential function.

22 healthy volunteers (25±4 years, 12 female, 10 male) were scanned on a 7T Philips Achieva MRI system. To optimize B₁ sensitivity within the VOI, a 15x15x1 cm³ high permittivity dielectric pad was placed between the volunteer's head and the 32-channel receive channel coil⁴. Fig. 1 shows the 8 cm³ volume of interest (VOI) planned in the a) mostly parietal WM (12 volunteers) and b) the mostly occipital GM (10 volunteers). The DWS data were acquired with a 13-interval STEAM sequence using bipolar diffusion gradients and cardiac synchronization⁵. Three orthogonal directions [1,1,-0.5], [1,-0.5,1], and [-0.5,1,1] were chosen to maximize the gradient strength for the isotropic diffusion weighting. DWS parameters were: TR/TE=3000/105 ms, Δ=100 ms, δ=30 ms, τ=13 ms, and 12 gradient amplitudes producing b-values of 0–17,204 s/mm². The isotropic spectra were analyzed using LCModel⁶ using an appropriate basis set and the DWS data were fitted to Eq. [1] using custom Matlab code.

Fig. 2 shows example diffusion-weighted spectra of total choline (tCho=choline+phosphocholine+ glycerophosphocholine), total creatine (tCr=creatine+phosphocreatine), and total N-acetyl-aspartate (tNAA=NAA+NAAglutamate). Fig. 3 shows example quantified LCModel values and fits to Eq. [1], demonstrating clear deviation from monoexponential decay on log-linear plots for both VOIs. To simplify results presentation, statistical significance is shown only when the null hypothesis could not be rejected in a standard unpaired two-tailed Students t-test (p<0.05), shown in Tables 1 and 2 for mean and 95% confidence interval estimations of D and α across all subjects for both VOIs.

Strikingly, the metabolites exhibit more non-Gaussian behavior as expressed by anomalous subdiffusion (lower α) from WM to GM, opposite compared to water trends (higher α). Furthermore, within the WM, although D was statistically different for each metabolite, α was indistinguishable (p=0.67) for tCho, tCr, and tNAA, indicating not only clear subdiffusion, but also that the axonal and glial compartments appear similar at high spatial resolution⁷. Within GM, although D was statistically different for each metabolite, α was indistinguishable between tCho and tNAA (p=0.17) and between tCr and tNAA (p=0.20). The trend that intracellular metabolites have lower values of D in GM compared to WM coincides with previous studies⁸.

Lower α values for tCho from WM to GM indicate that GM astrocyte intracellular space is more heterogeneous compared to WM. Furthermore, the change in α for tCho may be reflecting fibrous astrocytes in WM (overlapping, longer projections) and protoplasmic astrocytes in GM (shorter, thin branching projections)^8,9. Decreasing α values for tNAA from WM to GM may be indicative of confined NAA within the mitochondria, greater presence of neuronal organelles, and increased molecular crowding^10-12. As tCr is in both the protoplasmic astrocytes, cell bodies, axons, and dendrites, the decrease in α from WM to GM is consistent with tCho and tNAA. Finally, considering intracellular metabolites are more subdiffusive from WM to GM, but water diffusion is more Gaussian from WM (α~0.62) to GM (α~0.83), our results suggest the following possibilities: GM extracellular space is relatively more homogeneous and less hindered than WM extracellular space; and, intracellular/extracellular exchange is much faster in GM than WM (also reported using filtered exchange imaging¹⁴).

We have found that intracellular metabolites ubiquitously exhibit clear non-Gaussian subdiffusion in the healthy human brain, which may be a fundamental marker of biological cellular environments to be compact for molecular assembly and chemical reaction rate efficiency¹³.